Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -1,29 +1,67 @@
|
|
1 |
-
# app.py
|
2 |
-
import pandas as pd
|
3 |
-
from transformers import pipeline
|
4 |
-
import gradio as gr
|
5 |
|
6 |
-
# Load synthetic data
|
7 |
-
df = pd.read_csv("synthetic_profit.csv")
|
8 |
|
9 |
-
#
|
|
|
10 |
qa = pipeline(
|
11 |
"table-question-answering",
|
12 |
model="google/tapas-base-finetuned-sqa",
|
13 |
tokenizer="google/tapas-base-finetuned-sqa"
|
14 |
)
|
15 |
|
16 |
-
|
17 |
-
|
18 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
19 |
|
20 |
demo = gr.Interface(
|
21 |
fn=answer,
|
22 |
-
inputs=gr.Textbox(lines=2, placeholder="e.g.
|
23 |
-
outputs="
|
24 |
title="S/4HANA Profitability Chat",
|
25 |
-
description="Ask questions of synthetic S/4HANA data using TAPAS"
|
26 |
)
|
27 |
-
|
28 |
-
if __name__ == "__main__":
|
29 |
-
demo.launch()
|
|
|
|
|
|
|
|
|
|
|
1 |
|
|
|
|
|
2 |
|
3 |
+
# 3) load TAPAS
|
4 |
+
from transformers import pipeline
|
5 |
qa = pipeline(
|
6 |
"table-question-answering",
|
7 |
model="google/tapas-base-finetuned-sqa",
|
8 |
tokenizer="google/tapas-base-finetuned-sqa"
|
9 |
)
|
10 |
|
11 |
+
# 4) cast to strings to avoid the regex bug
|
12 |
+
df_str = df.astype(str)
|
13 |
+
|
14 |
+
# 5) sanity check
|
15 |
+
print( qa(table=df_str, query="What was the ProfitMargin for Product B in EMEA Q2 2024?") )
|
16 |
+
|
17 |
+
# 6) launch Gradio
|
18 |
+
import gradio as gr
|
19 |
+
|
20 |
+
import re
|
21 |
+
|
22 |
+
def answer(q: str) -> str:
|
23 |
+
# --- 1. try to parse explicit total/average queries ---
|
24 |
+
m = re.search(r"\b(total|average)\s+(ProfitMargin|Profit|Revenue|Cost)\b", q, re.IGNORECASE)
|
25 |
+
p = re.search(r"\bProduct\s*([A-D])\b", q, re.IGNORECASE)
|
26 |
+
t = re.search(r"\b(Q[1-4])\s*(\d{4})\b", q, re.IGNORECASE)
|
27 |
+
|
28 |
+
if m and p and t:
|
29 |
+
agg_type = m.group(1).lower() # "total" or "average"
|
30 |
+
metric = m.group(2) # column name
|
31 |
+
product = f"Product {p.group(1).upper()}"
|
32 |
+
quarter = t.group(1)
|
33 |
+
year = int(t.group(2))
|
34 |
+
|
35 |
+
# filter the *numeric* DataFrame
|
36 |
+
subset = df[
|
37 |
+
(df["Product"] == product) &
|
38 |
+
(df["FiscalQuarter"] == quarter) &
|
39 |
+
(df["FiscalYear"] == year)
|
40 |
+
]
|
41 |
+
|
42 |
+
if not subset.empty:
|
43 |
+
if agg_type == "total":
|
44 |
+
val = subset[metric].sum()
|
45 |
+
return f"Total {metric} for {product} in {quarter} {year}: {val:,.2f}"
|
46 |
+
else: # average
|
47 |
+
val = subset[metric].mean()
|
48 |
+
# show 3 decimal places for margins, 2 for currency
|
49 |
+
fmt = "{:,.3f}" if metric=="ProfitMargin" else "{:,.2f}"
|
50 |
+
return f"Average {metric} for {product} in {quarter} {year}: " + fmt.format(val)
|
51 |
+
|
52 |
+
# --- 2. fallback to TAPAS for everything else ---
|
53 |
+
res = qa(table=df_str, query=q)
|
54 |
+
agg = res.get("aggregator","")
|
55 |
+
if agg and agg != "NONE":
|
56 |
+
return f"Answer: {res['answer']} (agg: {agg})"
|
57 |
+
# last-resort: raw answer
|
58 |
+
return f"Answer: {res['answer']}"
|
59 |
+
|
60 |
|
61 |
demo = gr.Interface(
|
62 |
fn=answer,
|
63 |
+
inputs=gr.Textbox(lines=2, placeholder="e.g. Profit for Product A in Q1 2023?"),
|
64 |
+
outputs="text",
|
65 |
title="S/4HANA Profitability Chat",
|
|
|
66 |
)
|
67 |
+
demo.launch(share=True, debug=True)
|
|
|
|